Publication VII - Aalto

Publication VII

Ö. Bulakci, A. Bou Saleh, S. Redana, B. Raaf, J. Hämäläinen. Resource

Sharing in LTE-Advanced Relay Networks: Uplink System Performance Ana-

lysis. Transactions on Emerging Telecommunications Technologies (ETT),

[Invited Paper] Special Issue on Multi-Carrier Transmission, LTE and LTE-

Advanced, vol. 24, issue 1, pp. 32-48, January 2013.

c© 2013 John Wiley & Sons, Ltd.

Reprinted with permission.

171

TRANSACTIONS ON EMERGING TELECOMMUNICATIONS TECHNOLOGIESTrans. Emerging Tel. Tech. 2013; 24:32–48

Published online 28 August 2012 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/ett.2569

SPECIAL ISSUE - LTE-A

Resource sharing in LTE-Advanced relay networks:uplink system performance analysis†

Ömer Bulakci1,2*, Abdallah Bou Saleh2, Simone Redana1, Bernhard Raaf1 andJyri Hämäläinen2

1 Nokia Siemens Networks, NSN-Research, Radio Systems, St.-Martin-Str. 76, 81541, Munich, Germany2 Aalto University School of Electrical Engineering, Department of Communications and Networking, PO Box 13000, FI-00076Aalto, Finland

ABSTRACT

Relay-enhanced networks are expected to fulfil the demanding coverage and capacity requirements in a cost-efficient way.Type 1 inband relaying has been standardised as an integral part of the Third Generation Partnership Project (3GPP) Long-Term Evolution Release 10 and beyond (LTE-Advanced). This type of relay nodes (RNs) supports a relaying mode wherethe RN to donor evolved node B (donor eNB, DeNB) link (relay link, a.k.a. backhaul link) transmission is time-divisionmultiplexed with the RN-served user equipments (RUEs) to RN link (access link) transmission, whereas macrocell-serveduser equipments (MUEs) share the same resources with the RNs at DeNB. Hence, system performance depends stronglyon the resource sharing strategy among and within the links. Further, the set of subframes assigned for the relay linktransmission is semi-statically configured and thus a dynamic reconfiguration to adapt to fast-changing system conditions(e.g. RN cell load) is not viable. Besides, in order to fully exploit the benefits of relaying, the inter-cell interference, whichis increased because of the presence of RNs, should be limited via a proper power control (PC) scheme on each link.Therefore, an optimisation of both the resource sharing and PC strategy is required to enhance the overall performance ofrelay networks. In order to tackle these issues, we employ a statistic-based over-provisioned backhaul subframe allocationto be utilised for flexible co-scheduling of RNs and MUEs at the DeNB. In addition, we propose a combination of RNscheduling based on the number of RUEs and user throughput throttling achieving max–min fairness. Performance analy-sis of various resource sharing techniques along with PC optimisation is then carried out within the LTE-Advanced uplinkframework in urban and suburban scenarios. Comprehensive results show that the proposed schemes achieve significantthroughput gains and high system fairness with substantially increased flexibility in resource allocation. Copyright © 2012John Wiley & Sons, Ltd.

KEY WORDS

LTE-Advanced; relay deployments; power control; resource sharing; co-scheduling; uplink

*Correspondence

Ömer Bulakci, Nokia Siemens Networks, NSN-Research, Radio Systems, St.-Martin-Str. 76, 81541, Munich, Germany.E-mail: [email protected]†A previous version of this paper was presented in the 8th International Workshop on Multi-carrier System and Solutions (MC-SS2011), Herrsching, Germany.

Received 7 March 2012; Revised 6 June 2012; Accepted 12 July 2012

1. INTRODUCTION

The set of stringent requirements for future radio accessnetworks has triggered the embodiment of relay nodes(RNs) as part of the next-generation network infrastructure.Relaying is considered an integral part of Third Gene-ration Partnership Project (3GPP) Long-Term EvolutionRelease 10 and beyond (LTE-Advanced), which is recently

accorded the official designation of International MobileTelecommunications Advanced specified by the Inter-national Telecommunication Union-Radiocommunicationsector for Fourth-Generation networks.The motive behind choosing relaying as an enhance-

ment technology to current radio access networks hasbeen well elaborated in the literature. Namely, RNspromise to increase the network capacity [1] and to better

32 Copyright © 2012 John Wiley & Sons, Ltd.

Ö. Bulakci et al.

distribute resources in the cell, or alternatively, extendthe cell coverage area [1–3]. Relaying is regarded a cost-efficient technology as well [4].A self-backhauling Type 1 RN is a layer 3 decode-

and-forward RN, which controls its own cell; that is, ithas its own physical cell ID and includes functionalitiessuch as radio resource management, scheduling and hybridautomatic repeat request retransmission control. As illus-trated in Figure 1, the connection between RN and thecore network is carried out through the donor evolvednode B (donor eNB, DeNB) via a wireless backhaul link,which is referred to, herein, as the relay link. Moreover,the link between user equipment (UE) and its serving RNis referred to as the access link and that between UE and itsserving eNB, or DeNB, as the direct link. The RN utilisesthe same frequency band on both the access and relaylinks (inband operation), and the links are time-divisionmultiplexed to prevent interference between the two links(half-duplex operation). However, the transmissions on thedirect and access links can coexist; that is, the RNs andmacrocells utilise full frequency reuse.The set of backhaul subframes, which are referred

to, herein, Un subframes, (Figure 1) is configured semi-statically by the DeNB. During the Un subframes, theDeNB may or may not schedule its macrocell-served UEs(MUEs). One common assumption in the performanceevaluation of relay deployments is that Un subframes areexclusively assigned for the RN transmissions. However,such a hard resource split reduces the resource utilisa-tion efficiency at the DeNB, as load conditions in RNcells may change rapidly. That is, because of the semi-static nature of the Un subframe configuration, the num-ber of Un subframes cannot be changed dynamically,and thus, a fast enough adaptation to the load conditionsmay not be attained. Accordingly, it is important to inves-tigate resource sharing schemes among and within thevarious links.Furthermore, relay deployments will require a more

detailed dimensioning and planning than conventionalsingle-hop networks. RNs can create severe inter-cell inter-ference in particular when a large number of RNs aredeployed in a cell. In addition, because of various linkcharacteristics such as antenna gain and propagation loss,

the transmit power levels of UEs and RNs can be sig-nificantly different. This can also result in different inter-ference characteristics during regular subframes, whichare referred to, herein, Uu subframes, and Un subframes(Figure 1). Then, power control (PC) becomes an importantmeans in the uplink (UL) transmission not only to compen-sate for channel variations, but also to mitigate the interfer-ence and to limit the receiver dynamic range especially atDeNB. A high receiver dynamic range increases the sus-ceptibility of Single Carrier-Frequency Division MultipleAccess (SC-FDMA) to the loss of orthogonality, which cancause severe intra-cell interference [5]. The PC optimisa-tion methodology presented in our previous studies [6, 7]is taken as a basis in this work.Herein, we first present two resource sharing schemes

for the resource split at DeNB between RNs and MUEs,namely, hard resource split, where RNs are exclusivelyallocated all resources during Un subframes, and flexibleresource split, where Un subframes are over-provisionedand the MUEs and RNs are co-scheduled during thesesubframes. In the latter approach, the goal of achievingdynamic and efficient resource sharing is moved to thescheduler, which can instantaneously adapt to changes insystem conditions, rather than relying on semi-static Unsubframe allocation [8]. Besides, the number of Un sub-frames is chosen considering the statistical distribution ofthe fraction of RN-served UEs (RUEs) in the macrocell(i.e. the total load of RN cells). Moreover, in both schemesthe impact of the number of Un subframes on the accesslink capacity is taken into account considering the half-duplex operation.Second, we focus on resource sharing schemes on indi-

vidual links. In [9], time-division and frequency-divisionmultiplexing of relay and access link transmissions wereinvestigated on the downlink (DL) excluding the resourcesharing within the links. In [10], resource sharing on therelay link according to the buffer state at the RNs wasinvestigated for urban scenarios without applying a PCoptimisation, and the results including the suburban sce-narios are provided in [11]. In this work, we adopt themodel presented in [10, 11] and propose a combination ofrelay link scheduling based on the number of RUEs anda throughput (TP) throttling technique achieving max–min

Figure 1. Exemplified 4-relay-node (RN) deployment with depicted links. RUEs, RN-served user equipment; DeNB, donor evolvednode B.

Trans. Emerging Tel. Tech. 24:32–48 (2013) © 2012 John Wiley & Sons, Ltd.DOI: 10.1002/ett

33

Ö. Bulakci et al.

fairness (MMF) in the end-to-end two-hop communication.The scheme is referred to as hop-optimisation model. Theoptimisations are carried out in conjunction with a PC strat-egy to guarantee a proper system performance. Thoroughsystem-level simulations within the LTE-Advanced con-text show that the proposed optimisation strategies achievesignificant TP gains and high system fairness along withsubstantially increased flexibility in resource allocation.We note that herein optimisation refers to a parame-

ter tuning within the allowed ranges in the LTE-Advancedstandard and an adjustment of resource management strate-gies such that an appropriate trade-off between targetperformance metrics can be achieved. For instance, aPC optimisation strategy targeting cell capacity tunes theparameters in a way that the cell capacity is maximisedwithout degrading the cell coverage below a given level.The remainder of the paper is organised as follows.

Section 2 provides the background material including PCand the key features of the general framework. In Section 3,different resource sharing techniques are presented. Thesystem model and simulation assumptions are given inSection 4. In Section 5, detailed performance evaluationand analysis are carried out. Finally, Section 6 concludesthe paper.

2. BACKGROUNDAND DEFINITIONS

In this section, we first recall the framework of LTE ULtechnology. Then, fractional PC (FPC) scheme of LTERelease 8 is outlined. The key features of the work andthe definitions are presented next.

2.1. LTE uplink technology

LTE UL has adopted SC-FDMA [12] because of its lowpeak-to-average power ratio allowing low power consump-tion and high efficiency at the UE [13]. This propertyis particularly important for UE battery lifetime andpower amplifier design. The total bandwidth availablefor the UL transmission is defined by the multiplex-ing scheme used. In our work, we assume frequency-division duplex (FDD) mode, where the UL and DL areeach allocated exclusive 10-MHz transmission bandwidth.In the frequency domain, the whole UL bandwidth isdivided into subbands, which are called physical resourceblocks (PRBs). The PRB defines the resource allocationgranularity in LTE. As the system is operating on 10-MHz bandwidth, there are Mtotal D 48 PRBs avail-able for data transmission on the physical uplink-sharedchannel (PUSCH), and two PRBs are reserved to thephysical uplink control channel. The main difference ofSC-FDMA compared with OFDMA is the single-carrierconstraint, where only a set of adjacent PRBs can beallocated to a user.‡ Further, the users can be assigned a

‡This constraint, which is defined in LTE Release 8, is considered

herein for the sake of backward compatibility.

Figure 2. Frequency-division duplex (FDD) LTE-Advancedframe structure considering Type 1 relay nodes (RNs). Tx,transmission; Rx, reception; MUE, macrocell-served userequipment; DeNB, donor evolved node B; RUEs, RN-served

user equipment.

subset of Mtotal PRBs in each transmission time interval(TTI, a.k.a. subframe).

2.2. Uplink frame structure infrequency-division duplexLTE-Advanced networks

An LTE frame spans in total 10 ms and comprises 10subframes. Given that LTE-Advanced networks are oper-ated with full frequency reuse, MUEs and RUEs areserved simultaneously on the same frequency band andtime slots by DeNBs and RNs, respectively. Yet, consid-ering the resource allocation strategy defined for inbandType 1 RNs in [14], relay and access link transmissionsare time-division multiplexed. During the Un subframes,RUEs are not scheduled on the UL, thus experiencingtransmission gaps. An example frame structure is givenin Figure 2, where two subframes are reserved for therelay link. Moreover, a maximum of six Un subframes canbe semi-statically allocated for the relay link [14]. In addi-tion, MUEs can be optionally scheduled with RNs in Unsubframes. That is, Un subframes are either assigned exclu-sively to RNs or MUEs and RNs are co-scheduled duringUn subframes.

2.3. LTE compliant uplink open-looppower control

Thanks to the orthogonality of SC-FDMA, intra-cell inter-ference is not of main concern in LTE. The aim ofPC mechanisms is then to maximise the received powerof wanted signals through compensating the long-termchannel variations, while limiting the amount of theinter-cell interference generated. Nevertheless, the receiverdynamic range§ of eNBs, DeNBs and RNs should also be

§The receiver dynamic range is defined as the difference in dB between

the 5th and 95th percentiles of the cumulative distribution function

(CDF) of the total received power.

34 Trans. Emerging Tel. Tech. 24:32–48 (2013) © 2012 John Wiley & Sons, Ltd.DOI: 10.1002/ett

Ö. Bulakci et al.

adjusted via PC to avoid intra-cell interference, where ahigh dynamic range can cause the loss of orthogonality [5].To fulfil these objectives in LTE, FPC [15] is used for thePUSCH (see Section 2.1 for further details on this channel)to determine the UE transmit power. In this work, FPC isalso employed for the relay-specific PUSCH (R-PUSCH)which is the physical channel for the UL relay link datatransmission. Accordingly, the transmit power of a node �

(UE or RN) that employs open-loop PC is given in dBm as

P� DminfPmax; P0 C 10 � log10 M� C ˛ � Lg (1)

In this equation,

� Pmax is the maximum allowed transmit power, whichhas an upper limit of 23 dBm for UE power class 3[16] and 30 dBm for RN transmissions (optionally37 dBm for suburban scenarios) [14].

� P0 is the power offset comprising cell-specific andnode-specific components, and it is used for control-ling the received signal-to-noise-ratio target that canbe set from �126 dBm to Pmax with a step sizeof 1 dB.

� M� is the number of PRBs allocated to node �.� ˛ is a cell-specific path loss compensation factor thatcan be set to 0.0 and from 0.4 to 1.0 with a step sizeof 0.1 via 3-bit signalling.

� L is the DL path loss estimate calculated by the node.

Open-loop PC compensates slow channel variations(i.e. path loss changes including shadowing). If ˛ is setto one in Equation (1), the path loss is fully compensatedand the resulting scheme is called full compensation PC(FCPC). For a given P0 value, FCPC improves the cell-edge user performance at the cost of increased inter-cellinterference as a result of higher transmit power levels.Yet, the inter-cell interference can be reduced by setting˛ smaller than one, which increases the cell-centre per-formance, however, at the cost of penalising the cell-edgeperformance [17–19]. One of the important determinantsof the motivation of studying the applicability of the exist-ing FPC for the relay networks is the desired backwardcompatibility between LTE Release 8 and LTE-Advancedusers, that is, the legacy LTE Release 8 users, which can-not be updated with a new PC scheme should also supportrelay operations.As mentioned before, RNs have their own independent

cells, for example, from legacy LTE Release 8 UE per-spective, RNs are identical to eNBs [14]. Therefore, PCparameters for RUEs in RN cells, that is, ˛RUE and P RUE

0 ,are configured by RNs, and their transmit power levelsare determined taking into account the path loss on theaccess link between the serving RN and RUEs, and thenumber of allocated PRBs according to Equation (1). Sim-ilarly, during Uu subframes, the PC parameters for MUEsin macrocells, that is, ˛MUE and PMUE

0 (Uu), are set inde-pendently by the serving eNB or DeNB. Furthermore, forthe co-scheduled MUE and RNs, the DeNB can configuredifferent P0 values, namely, PMUE

0 (Un) and P RN0 .

2.4. Physical downlink controlchannel limitation

In practice, the maximum number of the UEs that can bescheduled in each TTI is limited by the possible num-ber of scheduling grants which are carried by the phys-ical DL control channel (PDCCH). A scheduling grantincludes dedicated user information which is necessaryto decode the data channel, e.g. data bandwidth alloca-tion, and modulation and coding scheme (MCS). Typically8-10 UEs can be scheduled per TTI because of PDCCHlimitation (or PDCCH blocking) [20]. In this work, thisnumber is set to 8. That is, if there are more than eight UEsaccessing to an eNB, a DeNB or an RN, a subset con-sisting of eight UEs will be scheduled in each TTI. Theunscheduled UEs in a given TTI will be scheduled inother TTIs in a way that resource fairness is achieved inthe time domain. Note that in reality, PDCCH limitationchanges from TTI to TTI depending on the used controlchannel elements by the served UEs. A cell-centre userwith better channel conditions requires a smaller number ofcontrol channel elements, whereas this number increases asthe channel conditions worsen towards the cell edge [21].We further note that we assume that, in the considered relayscenario, there is no limitation because of relay-specificPDCCH (R-PDCCH), that is, all RNs can be scheduled in aTTI, as the relay link experiences good channel conditionsand a small number of RNs are deployed, namely, 4-RNand 10-RN deployments are considered herein.

2.5. Power limitation and adaptivetransmission bandwidth

The maximum number of PRBs—denoted by Mmax—which can be assigned to a UEs depends on the differ-ence betweenPmax and the per-PRB power spectral density(PSD) of that UEs. The per-PRB PSD of a UEs can beobtained via the open-loop component of Equation (1) bysettingM�= 1 such that

PSDUE DminfPmax; P0 C ˛ � Lg (2)

The actual PSD is given by Equation (2) as longas the UEs is not driven into power limitation; other-wise, Pmax will be equivalently spread over the assignedPRBs resulting in decreased signal-to-interference-plus-noise ratio (SINR) per PRB. Such an assignment mayresult in outage, especially when the UEs is experienc-ing poor channel conditions. Consequently, linearMmax isobtained from the difference Pmax�PSDUE given in dB as

Mmax D round�100:1.Pmax�PSDUE/

�(3)

Hence, power-limited UEs will not be assigned moreresources than those they can afford. The unallocatedresources can then be better utilised by other UEs result-ing in more efficient bandwidth usage. This functionalityis called adaptive transmission bandwidth (ATB) [22]. It is


35

Ö. Bulakci et al.

worth mentioning that in contrast to [22] where floor oper-ator is used, we have adopted round operator. The aim is todecrease the quantisation errors caused by this operationand to further enhance the performance over the P0 setof interest. It is as well empirically justified via our sim-ulations. A detailed comparison of ATB with fixed trans-mission bandwidth (FTB) is given in [23]. Therein, it isshown that ATB is particularly advantageous for suburbanscenarios having large inter-site distance (ISD). The reasonis mainly twofold. First, because of large ISD, there area considerable number of power-limited UEs experienc-ing poor channel conditions. Hence, compared with FTB,ATB avoids a decrease in per-PRB PSD, which results inincreased SINR levels at low SINR CDF regime. Second,the resources that are not used by the power limited UEsare allocated to other UEs with better channel conditions.This in turn improves the mean UEs TP. Therefore, ATB isapplied only for suburban scenarios in this work.

3. RESOURCE SHARINGTECHNIQUES

The performance of relay networks depends significantlyon the resource allocation strategy, that is, on the balancebetween different links that compete for resources or thatmay act as bottlenecks for other consecutive links. In thissection, we first outline the schemes of resource split atDeNB, which mainly aim at determining the number ofUn subframes. Next, we present the different investigatedresource sharing techniques on the relay link along withthe TP throttling schemes on the access link.

3.1. Resource split at donor evolved node B

The number of Un subframes needs to be tuned properlyto find a balance in the resource share between relayand direct links. Moreover, its impact on the access linkcapacity should also be taken into account, as RUEs expe-rience transmission gaps during Un subframes, which mayrender the access link a bottleneck in the two-hop com-munications. In this context, we investigate the followingtwo schemes.

3.1.1. Hard resource split.

One common assumption in the literature is that Unsubframes are exclusively assigned for the relay link trans-missions, for example, [1, 2, 6, 10, 11]. An example ofhard resource split is illustrated in Figure 3(a) wherethe resource allocation at DeNB is depicted on the time-frequency plane. In this scheme, MUEs also experience thesame transmission gaps as RUEs. Crucially, such a splitimplies a reduction in the total amount of resources avail-able to MUEs and, thus, a performance degradation in casethe relay link resources are not fully utilised because of,for example, lower RN cell load. Consequently, given thesemi-static nature of the Un subframe configuration, the

f

tOver-provisioned Un Subframes

(b)

f

tUn Subframes

(a)

RN DeNB (Un) MUE DeNB (Un) MUE DeNB (Uu)

Uu Subframes

Uu Subframes

RN r(1 )cm M

*r RN rM M

*r rM M

Figure 3. Resource split at DeNB: (a) hard resource split; (b)flexible resource split. RN, relay node; DeNB, donor evolvednode B; Un subframe, backhaul subframe; Uu subframe, regular

subframe; MUE, macrocell-served user equipment.

hard resource split can reduce the resource utilisation effi-ciency. In this work, we take this approach as a benchmark.We note that the optimal number of Un subframes in such acase is chosen according to the long-term average demandof UEs, and hence, on a short-term basis, it is not optimal.

3.1.2. Flexible resource split.

In this scheme, we over-provision the Un subframessuch that MUEs can be scheduled during Un subframes toadapt to RN cell load. The number of resources assigned toMUE per Un subframe in cell ‘c’ is set as

mc D .1 � �RN/Mr; 0 6 mc 6 Mr (4)

where �RN is the resource share of RNs per Un subframeand Mr is the total number of relay link resources in aUn subframe. Further, �RN is determined in terms of �RN,the RUEs fraction in the cell, and Nb, the number of Unsubframes out of 10 as

�RN Dminf1; �RN10

Nbg with �RN D URN

Uc(5)

where URN and Uc are the total number of RUEs and thetotal number of all UEs in the cell, respectively. Note thatin Equation (5), Nb is put in denominator to determine theresource share per Un subframe. It can be seen that, for agiven Nb, the RN and MUE resource shares per Un sub-frame are, respectively, proportional to the total numberof RUEs URN and the total number of MUEs Uc � URN,provided that �RN 6 Nb=10. In particular, Nb=10 sets anupper bound on �RN below which resource shares of RNsand MUEs can be flexibly adapted to changes in systemconditions and when �RN > Nb=10 all the resources in aUn subframe will be assigned to RNs, see min operator inEquation (5). Thus, the selection of Nb yields a trade-off,where, for example, a larger Nb means higher flexibility inadapting to system conditions, whereas a reduced access


Ö. Bulakci et al.

link capacity, since RUEs experience transmission, gapsduring Un subframes.The over-provisioning of Un subframes is exemplified in

Figure 3(b). Besides, it is assumed that the total number ofresources in Un and Uu subframes is the same and equalto Mr; that is, no particular resource partitioning schemesare considered. Further, MUEs and RNs can be scheduledon different parts of the spectrum, but this is omitted forillustration purposes. In this example, the number of Unsubframes is over-provisioned from two (Figure 3(a)) tofour, and MUEs are scheduled with RNs in Un subframes(Figure 3(b)). It is worth noting that not all subframescan be configured as Un subframes due to indispensablesynchronisation and broadcast channels [24]; however, inFigure 3, the Un subframes are shown to be consecutivefor clarity.

3.2. Resource sharing techniques on therelay link

After determining the number of Un subframes accordingto one of the resource splits at DeNB, it is important toinvestigate how to allocate the available resources to thedifferent RNs. The schemes considered in this work areas follows.

3.2.1. Access instantaneous

throughput proportional.

In the access instantaneous TP-proportional (AIT-P)scheme, the resource share of an RN is determined pro-portionally to the total instantaneous TP achieved on theaccess link of this RN. Note that access instantaneous TPis defined as the TP achieved on the access link whena resource fair Round Robin (RR) scheduler is consid-ered and all resources are fully utilised. That is, RNs areassigned resources on a demand basis, where the RN withmore demand on the access link will be assigned moreresources on the relay link. In this manner, mij , the num-ber of resources assigned to UEs j served by RN i , can bewritten in terms of Ma, the total number of PRBs on theaccess link, and ui , the number of UEs attached to RN i ,as

mij D Ma

ui; j D 1; : : : ; ui ; i D 1; : : : ; N (6)

where N is the number of active RNs in the overlayingmacrocell. Note that Ma is assumed to be identical for allRNs in the system; that is, no particular resource parti-tioning schemes are considered. The number of resourcesallocated to RN i can then be obtained as

mi DPui

jD1

Pmij

kD1RijkPN

iD1

Pui

jD1

Pmij

kD1Rijk

M�r ; i D 1; : : : ; N

(7)whereM�

r is the total number of PRBs reserved for the RNtransmissions (Figure 3) and Rijk is defined as the per-PRB spectral efficiency on the access link of UEs j to its

serving RN i . Rijk is obtained via an approximation basedon Shannon’s capacity formula adjusted by two parametersBeff and Aeff, which are denoted by bandwidth and SINRefficiencies, respectively [25]. It is given as

Rijk D S �8<:

0; SINRijk < SINRminBW � �max; SINRijk > SINRmaxBW � Beff � log2.1 C Aeff � SINRijk/; else

(8)where BW is the bandwidth per PRB, �max is the maxi-mum spectral efficiency depending on the highest availableMCS for a given SINRmax. Besides, S accounts for LTEUL overhead, for example, because of reference signals.

3.2.2. Access user equipment proportional.

The resource shares in an access UE proportional (AUP)resource allocation scheme are defined according to theratio of ui , the number of UEs attached to RN i , to thetotal number of RUEs URN in the cell. Thus, the number ofresources assigned to RN i is set as

mi D ui

URNM�r ; i D 1; : : : ; N (9)

3.3. Throughput throttling techniques onthe access link

The end-to-end TP of the RUEs depends on the quali-ties of both the access and relay links. That is, the TPachieved on the better link will be throttled to accommo-date only whatever passes through the bottleneck. Accord-ingly, the end-to-end TP of a RUEs j served by an RN i isobtained as

TPeij Dmin�TPaij ;TPrij

�; j D 1; : : : ; ui ; i D 1; : : : ; N

(10)where the user capacity on the access link is defined asTPaij and that on the relay link as TP

rij . Herein, we con-

sider two TP throttling techniques, namely UEs instan-taneous TP proportional (UIT-P) and MMF. The formerdistributes the TP on an access link quality basis, thusallowing UEs with good access channel to achieve high TP,whereas the latter tries to prioritise the low TP regime in thesystem by maximising the TP of UEs suffering from badaccess channels.

3.3.1. User equipment instantaneous

throughput proportional.

In the UIT-P scheme, the end-to-end TP of a user isdetermined proportionally to its instantaneous access TP,given in terms of the spectral efficiency of the user’s accesslink. Following Equations (6) and (10), the end-to-end TPof UEs j attached to RN i is

TPeij Dmin

0@

mijXkD1

Rijk ;

Pmij

kD1RijkPui

jD1

Pmij

kD1Rijk

miXkD1

Rik

1A

j D 1; : : : ; ui ; i D 1; : : : ; N (11)


37

Ö. Bulakci et al.

where mi is determined according to the resource sharingtechnique utilised on the relay link and Rik is the per-PRBspectral efficiency of the RN i .

3.3.2. Max–min fairness.

In the MMF scheme, resources are allocated to usersin such a way that the users with worse link qualities areprioritised. Consequently, this will increase the end-to-endTPs at low TP regime at the expense of performance degra-dation at high TP regime. The user TP is then throttledaccording to the following algorithm analogously to thewater filling approach:

(1) Determine by means of Equations (6) and (8)

TPaij DmijXkD1

Rijk ; j D 1; : : : ; ui

(2) Sort the RUEs served by RN i in an ascending orderwith respect to TPaij

(3) Initialise j D 1 and TP�i D Pmi

kD1Rik for RN i

(4) ifTP�

iui�jC1 > TPaij� TPeij D TPaij� TP�i D TP�i � TPeij

else� TPe

ijD TP�

iui�jC1 ; j D j ; : : : ; ui

� exit(5) while j < ui increment j D j C 1 and go to step 5.

4. SYSTEM MODEL

The simulated network is represented by a regular hexago-nal cellular layout with 19 tri-sectored sites (i.e. 57 cells).RNs are regularly deployed at the sector borders. Figure 4presents deployments of 4 RNs and 10 RNs per sector.Indoor users are assumed, where 25 uniformly distributedUEs are dropped per sector and the full buffer traffic modelis applied. In total, 200 user drops (or snapshots) are simu-lated using a Matlab-based system-level semi-static sim-ulator where statistics are collected from the inner-most

sector only to ensure proper modelling of interference.Simulation parameters follow the latest parameter settingsagreed in 3GPP [14] and are summarised in Table I.A frequency reuse factor of one (full reuse scheme) is

considered in the network. All available resources in a cellare assumed to be used at all times provided that UEsare not power limited (Section 2.5), and hence, a ratherpessimistic interference modelling is considered on theaccess link. Relay site planning is assumed as modelledin [14]. Directional antennas are utilised at the RNs forrelay link transmission, whereas omni-directional antennasare assumed for the access link transmission. Log-normalshadow fading is modelled for non-line-of-sight propaga-tion conditions, whereas fast fading is not simulated.In addition, it is assumed that the same Un subframe

configuration is utilised for all RNs in the network, and it isaligned by the operation and maintenance (O&M) system.Therefore, no RN-to-RN interference is considered. Notethat RN-to-RN interference [26, 27] occurs when an RNtransmits on its relay link and interferes with a receptionon the access link of another RN.

5. PERFORMANCE EVALUATIONAND ANALYSIS

Comprehensive system-level simulations have been carriedout to evaluate the performance of the different resourcesharing techniques within the LTE-Advanced framework.Moreover, 3GPP urban (case 1) and suburban (case 3)scenarios with ISDs of 500 and 1732 m, respectively, areconsidered [14]. For each scenario, deployments with 4and 10 RNs per cell are investigated. Furthermore, theeNB-only deployment with RR scheduler is taken as areference for performance comparisons. We note that aresource fair RR scheduler is utilised in the benchmark sothat the achieved relative gains through the studied tech-niques can be more easily clarified. With a more advancedscheduler in the benchmark, for example, one that priori-tises the cell-edge user performance, a trade-off could beattained where the performance is increased at the low TPregime at the cost of a reduction in the cell TP. Herein, this

(a) (b)

Figure 4. Relay node (RN) deployments at the cell border: (a) 4 RNs and (b) 10 RNs. DeNB, donor evolved node B; ISD,inter-site distance.


Ö. Bulakci et al.

Table I. Simulation parameters.

Parameter Value

System parameters

Carrier frequency 2 GHzBandwidth 10 MHzNumber of PRBs 48 for data + 2 for

control channelPRB bandwidth 180 kHzHighest MCS 64-QAM, R = 9/10

(�max=5.4 bps/Hz)Penetration loss 20 dB on direct and

access linksBandwidth efficiency .Beff/ 0.88SINR efficiency .Aeff/ 1/1.25Overhead scaling (S) 0.75Thermal noise PSD �174 dBm/HzSINR lower bound �7 dBeNB/DeNB parameters

Transmit power 46 dBmAntenna gain 14 dBiAntenna configuration Tx-2, Rx-2Noise figure 5 dBAntenna pattern A(� ) = �min Œ12.�=�3dB/

2;Am�

(horizontal) �3dB=70ı and Am = 25 dB

UE parameters

Maximum transmit power 23 dBmAntenna configuration Tx-1, Rx-2Antenna gain 0 dBiNoise figure 9 dB

RN parameters

Relay site planning Considered via higherprobability of LOS and5 dB bonus on NLOS

Transmit power 30 dBmAntenna configuration Tx-2, Rx-2RN-DeNB antenna gain 7 dBi(directional antenna)RN-UE antenna gain 5 dBiRelay link antenna pattern A(� ) = �minŒ12.�=�3dB/

2;Am�

(horizontal) �3dB=70ı and Am = 20 dBAccess link antenna pattern Omni-directionalNoise figure 5 dB

Shadowing

Shadow fading Log-normalStandard deviation 8 dB on the direct link

10 dB on the access link6 dB on the relay link

De-correlation distance 50 mCorrelation factor 0.5 between sites

1.0 between sectors

PRB, physical resource block; MCS, modulation and cod-ing scheme; SINR, signal-to-interference-plus-noise ratio; PSD,power spectral density; eNB, evolved node B; DeNB, donorevolved node B; UEs, user equipment; RN, relay node; LOS,line-of-sight; NLOS, non-line-of-sight; QAM, quadrature ampli-tude modulation.

would then translate into decreased relative gains at the lowTP regime, whereas the relative gains in terms of cell TPwould increase.Performance evaluation is carried out in terms of the

5th percentile (5%-ile) and 50th percentile (50%-ile) UEsTP CDF levels, and the cell capacity (aggregate user TP,a.k.a. cell TP), which mainly reflect cell-edge, medianand cell-centre user performances, respectively. Theseperformance metrics are conventionally utilised in Inter-national Mobile Telecommunications Advanced standard-isations (e.g. [14]). Moreover, the 50th percentile UEs TPCDF level is particularly critical for performance evalua-tions in suburban scenarios to assure an optimum perfor-mance enhancement [7]. This is due to high inhomogeneityof user experience within the cell, which implies substan-tially different UEs TP CDF levels. In our work, we focuson techniques to improve the low TP regime to achieve amore homogeneous user experience over the cell area andthus a high level of fairness in the system. Jain’s fairnessindex [28] is considered as the criterion to reflect the sys-tem fairness. An index of one indicates full fairness whereall UEs achieve the same TP.Jain’s fairness index is defined as in Equation (12) to

reflect fairness among RUEs, referred to as JFRN, and asJFsys in Equation (13) to reveal the fairness in the system asa whole. Recall that N is the total number of RNs, URN isthe total number of RUEs, ui is the number of UEs servedby RN i , and TPeij is the end-to-end TP of user j in RNcell i . UD is defined as the total number of MUEs servedon the direct link and TPD;l is the achieved TP of MUE l .

JFRN D

0@

NXiD1

uiXjD1

TPeij

1A

2

URN

NXiD1

uiXjD1

TPeij2

(12)

JFsys D

0@

NXiD1

uiXjD1

TPeij CUDXlD1

TPD;l

1A

2

.URN C UD/

0@

NXiD1

uiXjD1

TPeij2 C

UDXlD1

TP2D;l

1A

(13)In the remainder of this section, we first analyse

the resource sharing techniques under the hard resourcesplit assumption at DeNB, and then, further perfor-mance enhancements are demonstrated utilising the flexi-ble resource split at DeNB. All resource sharing techniquesare associated with a proper UL PC optimisation to guar-antee an overall system performance enhancement. Fur-thermore, in what follows, we define our reference modelto utilise AIT-P scheduling of RNs on the relay link andUIT-P TP throttling on the access link. Recall that we referto our proposed scheme, where AUP scheduling on therelay link along with MMF TP throttling on the access linkis utilised, as the hop-optimisation model.


39

Ö. Bulakci et al.

5.1. Resource sharing based on hardresource split at DeNB

The PC optimisation methodology presented in [6, 7] isfollowed. Accordingly, in urban scenarios, two strategiesare considered for the UEs UL transmissions, namely cellcoverage-oriented, that is, the 5th percentile UEs TP is pri-oritised, and the 50th percentile UEs TP-oriented settings.On one hand, the cell coverage-oriented setting employsFCPC (˛ D 1:0) and yields a higher 5th percentile UEsTP gain at the expense of reduced UEs TP gain for highpercentiles. On the other hand, the 50th percentile UEs TP-oriented setting employs FPC (˛ D 0:6) and yields a higherUEs TP gain for high percentiles at the expense of reducedUEs TP gain for low percentiles. In addition, for the relaylink, the PC parameters are set such that a higher capacitycan be attained. It is worth noting that the relay link capac-ity is highly interference limited, and thus, a fine-tuning ofPC parameters does not provide notable gains above a cer-tain level of P0 value for the relay link transmissions. Wefurther note that the cell capacity-oriented setting foundin [6, 7], that is, P0 D �55 dBm and FPC (˛ D 0:6),is applied for the eNB-only deployment. Furthermore, insuburban scenarios, the trade-off setting presented in [7]is considered for both UEs and RN transmissions and, aswell, for the eNB-only deployment. This setting is shownto provide a good balance between low and high UEs TPCDF levels [7]. These optimised PC parameter settings arepresented in Table II.

5.1.1. Selection of the number of

backhaul subframes.We proceed with determining the number of Un sub-

frames (Nb) yielding the optimum overall system perfor-mance. Figure 5 presents the UEs TP CDFs for differentUn subframe configurations in a 4-RN urban scenario. Inthis figure, we also illustrate a novel, very useful concept inperformance analysis of heterogeneous networks, which iscalled proportional CDFs. Namely, the CDFs correspond-ing to RUEs and MUEs are scaled by the ratio of these

0 200 400 600 800 1000 12000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

UE TP (kbps)

CD

F

eNB−only1 Un TTI: System2 Un TTIs: System4 Un TTIs: System1 Un TTI: MUEs4 Un TTIs: MUEs1 Un TTI: RUEs4 Un TTIs: RUEs

Figure 5. User equipment (UEs) throughput (TP) cumula-tive distribution functions (CDFs) considering different back-haul subframe (Un subframe) configurations in 4-relay-node(RN) deployment; urban scenario. MUEs, macrocell-servedUEs; RUEs, RN-served UEs; eNB, evolved node B; TTI,

transmission time interval.

UEs to the total number of UEs in the system to obtainthe proportional CDFs. Accordingly, the system UEs TPCDF is the sum of both. Equivalently, because we definethe RN coverage as the ratio of �RN D Ef�RNg, where�RN is defined in Equation (5) and Ef�g denotes statis-tical expectation, and the user-drop is uniform, the scal-ing factors of the proportional CDFs of RUE and MUEsare given as �RN and 1 � �RN, respectively. The propor-tional CDFs are of particular interest in relay deploymentsbecause they clearly reflect the impact of various optimi-sation techniques on the individual performances of RUEsand MUEs, and hence, reveal the reason of the resultantimpact on the system performance.In the example scenario, AIT-P scheduling and UIT-

P TP throttling were used on the relay and accesslinks, respectively, employing the 50th percentile UEs TP-oriented PC setting. It is seen in Figure 5 that for the casewith insufficient number of Un subframes (cf. 1 Un TTI),the system performance is mainly limited by the perfor-mance of RUEs because of the Un bottleneck, whereas the

Table II. Optimised power control parameter configurations for relay nodes (hard resource split at donorevolved node B), macrocell-served user equipments (Uu) and relay node-served user equipments.

Urban Scenarios SuburbanScenarios

Cell coverage- 50th percentile UE TP- Trade-offoriented setting oriented setting setting

MUE RUE RN MUE RUE RN MUE RUE RN(Uu) (Uu) (Uu)

P0 �95 �101 �83 �51 �61 �55 �63 �63 �63[dBm]˛ 1.0 1.0 1.0 0.6 0.6 0.6 0.6 0.6 0.6Pmax 23 15 30 23 15 30 23 15 30[dBm]

UE, user equipment; TP, throughput; RN, relay node; MUE, macrocell-served user equipment; RUE,RN-served user equipment; (Uu), during Uu subframes; (Un), during Un subframes.


Ö. Bulakci et al.

MUEs performance becomes the limitation when four Unsubframes are used, as only six subframes remain avail-able for the MUEs. In this case, the RUEs performanceis significantly increased resulting in a higher TP gain athigh percentiles. Besides, it is seen that the case with twoUn subframes achieves the best gain at the low TP regime.As the emphasis in LTE-Advanced is a more homogeneoususer experience, the optimum number of Un subframes ischosen to be two. Similarly, the optimum number of Unsubframes is found for other scenarios, and these settingsare given in Table III along with the RN coverage areas(i.e. �RN).

5.1.2. Performance assessment.

Assuming the hard resource split at DeNB according toSections 3.1.1 and 5.1.1, we first evaluate the performanceof the different resource sharing strategies and TP throt-tling techniques presented in Sections 3.2 and 3.3. It isworth noting that these techniques will only shape the per-formance of RUEs, and thus, the performance of MUEswill not be impacted.The simulation results for urban scenarios are shown

in Figure 6(a). The percentage gains at depicted per-centiles of UEs TP CDFs are presented with respect tothe eNB-only deployment in Figure 6(a) for the 4-RN and10-RN deployments. In the 4-RN deployment with the50th percentile UEs TP-oriented PC, it is observed thatthe reference and hop-optimisation models yield, respec-tively, 63% and 118% TP gains at the 5th percentile UEsTP CDF, and 37% and 47% TP gains at the 50th per-centile UEs TP CDF. That is, the hop-optimisation modelachieves significant TP gains at the 5th percentile UEsTP CDF (extra 55%) and moderate gains at the 50th per-centile UEs TP CDF (extra 10%) without degrading thecell capacity (cf. Cell TP) compared with the referencemodel. The 5th percentile TP gain can be further increasedat the expense of somewhat reduced the 50th percentileTP gain and a slight degradation in cell capacity, whenthe cell coverage-oriented PC is utilised. A similar per-formance enhancement is also observed in the 10-RNdeployment. Nevertheless, compared with the case in the4-RN deployment, it is seen that the hop-optimisationmodel yields higher relative gains at the 50th percentileUEs TP CDF over the reference model. Furthermore, the

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Jain’s Fairness Index

CD

F

eNB−only & Cell Capacity−oriented PCReference & PC 1: SystemReference & PC 1: RUEsHop−Optimization & PC 2: SystemHop−Optimization & PC 2: RUEsHop−Optimization & PC 1: SystemHop−Optimization & PC 1: RUEs

PC 1: 50%−ile UE TP−oriented PCPC 2: Cell Coverage−oriented PC

(a)

(b)

Cell TP50%-ile TP5%-ile TP0%

50%

100%

150%

200%

Figure 6. (a) Throughput (TP) percentage gains over evolvednode B (eNB)-only in 4-relay-node (RN) and 10-RN deployments;(b) Jain’s fairness index in 10-RN deployment; urban scenario.UEs, user equipment; PC, power control; RUEs, RN-served UEs;

CDF, cumulative distribution function.

CDFs based on the Jain’s fairness index are plotted inFigure 6(b) for the 10-RN deployment. It is observedthat the hop-optimisation model results in significantlyhigher fairness in the system compared with both eNB-only deployment and the reference model. In addition, thesystem fairness can be further enhanced using the cellcoverage-oriented PC scheme. Besides, the mean Jain’sfairness indices for different deployments are tabulated inTable IV. Therein, the 50th percentile UEs TP-oriented PCand cell coverage-oriented PC settings are depicted by PC1 and PC 2, respectively.

Table III. Optimum backhaul subframe configurations for different resource splitsat donor evolved node B.

Deployment scenario RN coverage Number of Unarea (%) subframes

Hard split Flexible split

Urban—4 RNs 29.5 2 4Urban—10 RNs 45.5 4 6Suburban—4 RNs 43.5 4 6Suburban—10 RNs 67.0 6 6

RN, relay node.


41

Ö. Bulakci et al.

Table IV. Mean Jain’s fairness indices for the investigated scenarios.

Deployment Scenario Mean Fairness Index Mean System Fairnessfor RUEs Index

Ref. Hop-opt. Model Ref. Hop-opt. Model

model PC 1 PC 2 model PC 1 PC 2

Urban—4 RNs 0.7330 0.9833 0.9833 0.7358 0.8403 0.8680Urban—10 RNs 0.6932 0.9728 0.9716 0.6702 0.8575 0.8657Suburban—4 RNs 0.7374 0.9757 0.6347 0.7710Suburban—10 RNs 0.6790 0.9709 0.7045 0.8761

RN, relay node; RUEs, RN-served user equipment; PC, power control.

The proposed hop-optimisation model achieves, as well,significant TP gains in suburban scenarios as presentedin Figure 7(a). We first note the significant performanceenhancement at the 5th percentile UEs TP CDF overthe eNB-only deployment after RNs are deployed. Thisbehaviour shows that RN deployments can effectively copewith the coverage issues due to large ISD. In the 4-RNdeployment, the reference and hop-optimisation modelsachieve, respectively, 297% and 310% TP gains at the 5thpercentile UEs TP CDF, and 95% and 136% TP gains at the50th percentile UEs TP CDF. That is, the hop-optimisation

(a)

(b)

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Jain’s Fairness Index

CD

F

eNB−only & Trade−off PCReference & Trade−off PC: SystemReference & Trade−off PC: RUEsHop−Optimization & Trade−off PC: SystemHop−Optimization & Trade−off PC: RUEs

Cell TP50%-ile TP5%-ile TP0%

500%

1000%

1500%

2000%

Figure 7. (a) Throughput (TP) percentage gains over evolvednode B (eNB)-only in 4-relay-node (RN) and 10-RN deployments;(b) Jain’s fairness index in 10-RN deployment (b); suburban sce-nario. PC, power control; RUEs, RN-served user equipment;

CDF, cumulative distribution function.

model outperforms the reference model both at the 5thpercentile and 50th percentile UEs TP CDF levels (seeextra gains depicted by � in Figure 7(a)). Lower gainat the 5th percentile UEs TP CDF is due to the lim-itation of the cell-edge MUEs. On the other hand, inthe 10-RN deployment, the gains for the reference andhop-optimisation models read, respectively, as 1090%and 1664% at the 5th percentile UEs TP CDF, and 142%and 196% at the 50th percentile UEs TP CDF. This impliesthat the hop-optimisation model significantly outperformsthe reference model both at the 5th percentile and 50thpercentile UEs TP CDF levels (see extra gains depictedby � in Figure 7(a)). Furthermore, in both 4-RN and 10-RN deployments a negligible loss is observed in the cellcapacity when the proposed model is used. In addition,according to the fairness CDFs plotted in Figure 7(b) forthe 10-RN deployment, it is seen that the hop-optimisationmodel results in significantly higher fairness in the systemcompared with both eNB-only deployment and the refer-ence model. Recall that the mean Jain’s fairness indices fordifferent deployments are tabulated in Table IV.

5.2. Resource sharing based on flexibleresource split at DeNB

The preceding analysis has shown that the proposed hop-optimisation model achieves significantly higher gainscompared with the reference model in all of the consideredscenarios. Therefore, we consider the hop-optimisationmodel in the following and further show the impact of theflexible resource split scheme.The optimisation strategies in case of the flexible

resource split scheme are more involved. The reason isthreefold. First, as mentioned before, interference charac-teristics during Uu and Un subframes can be substantiallydifferent, which implies different PC parameters for theco-scheduled MUEs and further tuning of those pertain-ing to RNs during Un subframes. Second, co-scheduledMUEs and RNs may have significantly different receivedpower levels at the serving DeNB, and hence, the receiverdynamic range of the serving DeNB should also be takeninto account. Third, the choice of MUEs selection schemefor co-scheduling can impact the resultant system perfor-mance drastically. Nevertheless, the possibility of MUEs


Ö. Bulakci et al.

selection for co-scheduling increases the degree of free-dom in optimisation strategies where various trade-offs canbe attained.

5.2.1. Selection of the number of

backhaul subframes.

In line with [8], the over-provisioning of Un subframesaims at adapting up to percentile of F .�RN/, which is theCDF of �RN. The resultant number of Un subframes, Nb,is then determined taking the upper bound �RN 6 Nb=10

(Section 3.1.2) into account as

Nb D round.10 � F�1.// with 0 < Nb 6 6 ; Nb 2 NC(14)

With the D 90th percentile target,Nb is determined fordifferent scenarios. In Figure 8, CDFs of �RN are plottedfor the considered scenarios. It is seen that the D 90thpercentile CDF level corresponds to �RN = F�1.0:9/ �f0:42; 0:59; 0:57; 0:79g for fUrban 4-RN, Urban 10-RN,Suburban 4-RN, Suburban 10-RNg deployments, respec-tively. Following Equation (14), Nb values are thenfound and tabulated in Table III for the investigatedscenarios. However, because of constraints on Nb given inEquation (14), the target level cannot be achievedfor the considered scenarios. Accordingly, achievable tar-get CDF levels, denoted by � D F .Nb=10/, aredepicted in Figure 8. We recall that � values markthe levels up to which the resource split can be flex-ibly tuned. It is worth noting that for Urban 4-RNdeployment, the flexibility of the scheme can be fur-ther increased for Nb > 4; however, for such values,access link limitation is observed because of increasednumber of transmission gaps, and thus, Nb > 4 isnot preferred.

5.2.2. Performance assessment.

In this section, the performance is analysed consider-ing random MUEs selection for co-scheduling. We thenstart with illustrating PC optimisation strategy during Un

0 10 20 30 40 50 60 70 80 90 1000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

RN(%)

CD

F

UrbanScenario

SuburbanScenario

Case 1: 4 RNsCase 1: 10 RNsCase 3: 4 RNsCase 3: 10 RNs

λ*=21%−ile

λ*=92%−ile

λ*=96%−ile

λ*=87%−ile

Figure 8. The cumulative distribution functions (CDFs) of relaynode (RN)-served user equipment (UE) fraction (�RN) for urban

and suburban scenarios.

subframes, taking TP gains and DeNB receiver dynamicrange into account. In Figure 9, the results for the 10-RNdeployment in urban scenario are provided. Therein, PCoptimisation during the Un subframes is depicted whereFPC is utilised, and a parameter sweep is applied for theP0 values of RNs (P RN

0 ) at two different P0 values ofco-scheduled MUEs (PMUE

0 ). During Uu subframes, the50th percentile UEs TP-oriented setting is used (Table II).The percentage gains at the 50th percentile UEs TP CDFover eNB-only deployment (left vertical axis) and DeNBreceiver dynamic ranges (right vertical axis) are presented.Two settings are marked, which yield comparable resultsin terms of the shown performance measures, that is, themaximum 50th percentile UEs TP CDF and the minimumDeNB receiver dynamic range. Besides, these settings alsoachieve similar percentage gains at the 5th percentile UEsTP CDF level over eNB-only deployment (around 205%)and mean UEs TP values (around 850 kbps). Neverthe-less, the second setting is selected because the PMUE

0values during Uu and Un subframes become the sameso that no extra signalling is required to update the P0 val-ues for Un subframes. The corresponding P0 values alongwith the achieved gains over eNB-only deployment areprovided in Table V (Random MUEs Selection Scheme).Because of the full frequency of reuse, the total inter-cell

interference at DeNB is of particular concern for the resul-tant system performance. During each of the Uu and Unsubframes, we group all interferers into two distinct groupsdenoted by �1 and �2. To be specific, �1 and �2 corre-spond to, respectively, the MUEs in other cells and RUEsin all cells during Uu subframes and the co-scheduledMUEs in other cells and RNs in other cells during Un sub-frames. The total instantaneous interference-over-thermal(IoT) level per PRB at DeNB can then be written in lineardomain as

IoTDeNB D I C PN

PND I�1

C I�2C PN

PN(15)

−6370

75

80

85

90

95

50%

−ile

UE

TP

Gai

n ov

er e

NB

−on

ly [%

]

P0

RN [dBm]

−59 −55 −51 −478

10

12

14

16

18

Gain:P =−47dBm

Gain:P =−51dBm

Dyn. R.:P =−47dBm

Dyn. R.:P =−51dBm

FPC & 10 RNs

2 1

2 1

Rec

eive

r D

ynam

ic R

ange

at D

eNB

[dB

]

Figure 9. Power control (PC) optimisation based on fractionalPC (FPC) during backhaul subframes (Un subframes) in 10-relay-node (RN) deployment; urban scenario. Dyn. R, receiverdynamic range; TP, throughput; eNB, evolved node B; DeNB,

donor eNB.


43

Ö. Bulakci et al.

Table V. Performance of flexible resource split at donor evolved node B in urban scenarios.

MUE Scenario PC TP Gain (%) P0 (dBm)Selection Scheme over eNB-only

Scheme 90th 50th 5th MUE MUE RNpercentile percentile percentile (Uu) (Un)

Random 4 RNs FPC 39 63 125 �51 �51 �59FCPC 21 47 161 �95 �79 �83

10 RNs FPC 48 92 205 �51 �51 �59FCPC 31 83 238 �95 �79 �83

Cell 4 RNs FPC 45 68 85 �51 �51 �59centre FCPC 37 53 103 �95 �79 �83

10 RNs FPC 54 101 103 �51 �51 �59FCPC 46 91 116 �95 �79 �83

Cell 4 RNs FPC 41 73 83 �47 �51 �55edge FCPC 22 61 126 �91 �83 �83

10 RNs FPC 51 101 189 �47 �51 �55FCPC 32 91 219 �91 �83 �83

MUE, macrocell-served user equipment; RN, relay node; PC, power control; FPC, fractional PC; FCPC,full compensation PC; TP, throughput; eNB, evolved node B; (Uu), during Uu subframes; (Un), duringUn subframes.

where I�1and I�2

are the interferences caused by nodegroups �1 and �2, respectively, and PN is the thermal noisepower. Equation (15) can be rewritten as

IoTDeNB D I�1C PN

PNC I�2

C PN

PN� 1

D IoT�1C IoT�2

� 1

(16)

Applying statistical expectation Ef ;t f�g over frequencyand time to the both sides of Equation (16), the mean IoTat DeNB can be formulated as

IoTDeNB D IoT�1C IoT�2

� 1 (17)

Subsequently, to highlight the difference in interferencecharacteristics during Uu and Un subframes, we plot meanIoT CDFs in Figure 10. It is first seen that the IoT pro-portion of the MUEs is dominating the total mean IoTat DeNB. Second, it is observed that the total mean IoTat DeNB tends to be less during Un subframes than thatduring Uu subframes. Another observation during Un sub-frames is that the mean IoT levels from RNs are compa-rable with or less than those of the MUEs. The reasonis stemming from the low transmit power levels of RNs,thanks to decreased path loss towards the serving DeNBdue to relay site planning, and higher antenna gains, aswell as the directional antennas installed at RNs for back-hauling. We note that this reveals a key advantage of co-scheduling. Namely, because of the decreased total meanIoT at DeNB during Un subframes, the experienced SINRvalues for the co-scheduled MUEs are substantially higherthan those during Uu subframes provided that a proper PCsetting is applied.

5 10 15 20 25 300

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

CD

F

@ Un

@ Uu

Figure 10. Mean interference-over-thermal (IoT) cumulativedistribution functions (CDFs) during regular subframes (Uusubframes) and backhaul subframes (Un subframes) in 10-relay-node (RN) deployment; urban scenario. MUEs, macrocell-served user equipment; RUEs, RN-served user equipment.

Similar optimisation is performed also for other urbanscenarios. The corresponding UEs TP CDFs are plotted inFigure 11(a) for 4-RN deployment and in Figure 11(b) for10-RN deployment. Moreover, the P0 values along withthe achieved gains over eNB-only deployment are pro-vided in Table V (Random MUEs Selection Scheme). For4-RN deployment, it is observed that the flexible resourcesplit scheme yields high gain at all percentiles up to 98thpercentile CDF level over the hard resource split scheme.Slight performance degradation is noticed only at very highpercentiles. Furthermore, FPC and FCPC enable a trade-off between low and high TP regimes. In case of 10-RNdeployment, it is first seen that by deploying more RNs, the


Ö. Bulakci et al.

0 500 1000 15000

0.10.20.30.40.50.60.70.80.9

1

UE TP (kbps)

(a)

(b)

CD

F

eNB−onlyHard Resource Split: FCPCFlexible Resource Split: FCPCHard Resource Split: FPCFlexible Resource Split: FPC

4 RNs

0 500 1000 15000

0.10.20.30.40.50.60.70.80.9

1

UE TP (kbps)

CD

F

eNB−onlyHard Resource Split: FCPCFlexible Resource Split: FCPCHard Resource Split: FPCFlexible Resource Split: FPC

10 RNs

Figure 11. User equipment (UEs) throughput (TP) cumulativedistribution functions (CDFs) in (a) 4-relay-node (RN) deploy-ment and (b) 10-RN deployment; urban scenario. FPC, fractionalpower control; FCPC, full compensation power control; eNB,

evolved node B.

system performance can be further enhanced. Moreover,the flexible resource split scheme results in higher TP gainsin low TP regime compared with the case of 4-RN deploy-ment. Another crucial observation for the FCPC scheme isthat the PMUE

0 value in Un subframes is 16 dB higher thanthat in Uu subframes to cope with different interferencecharacteristics. To be specific, using the PMUE

0 value foundfor Uu subframes also in Un subframes results in drasticperformance degradation for the co-scheduled MUEs andthus a highly inhomogeneous TP distribution.A similar optimisation strategy is followed for suburban

scenarios. The UEs TP CDFs are shown in Figure 12 wherethe corresponding P0 values along with the achieved gainsrelative to eNB-only deployment are presented in Table VI(Random MUEs Selection Scheme). We note that unlessotherwise stated, the PC parameters of a node are keptthe same as given in Table II. In both 4-RN and 10-RNdeployments, the flexible resource split outperforms thehard resource split scheme up to the 94th percentileCDF level.

5.2.3. Impact of macrocell-served user

equipment selection schemes.

Different strategies can be employed for the MUEsselection such as the link quality. Herein, we exemplify

0 200 400 600 800 1000 1200 1400 1600 1800 20000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

UE TP (kbps)

CD

F

4 RNs

10 RNs

Figure 12. User equipment (UE) throughput (TP) cumulativedistribution functions (CDFs) in 4-relay-node (RN) and 10-RNdeployments; suburban scenario. Res., resource; Sel., selection;

eNB, evolved node B; MUEs, macrocell-served UE.

two MUEs selection schemes. In the example schemes,cell-edge or cell-centre MUEs are selected to be sched-uled with RNs in Un subframes. In order to classify theMUEs as cell-edge UEs, the periodic measurements of ref-erence signal received power (RSRP) and reference signalreceived quality (RSRQ) can be used [29].Similarly, in the considered example schemes, reference

signal received power measurements are employed to clas-sify the MUEs as cell edge or cell centre. Specifically, foreach MUEs the numbers of neighbour macrocells and RNcells are determined such that the DL signal strength ofthe neighbour macrocell or RN cell is 6 dB lower than theDL signal strength of the serving cell. The MUEs withthe largest number of such neighbours, where the num-ber pertaining to macrocells is prioritised, are consideredas cell-edge MUEs, and the rest of the MUEs are consid-ered as cell-centre UEs. For example, assume that MUEs1, which has two such macrocell and three RN cell neigh-bours, and MUEs 2, which has three such macrocell andtwo RN cell neighbours. According to the aforementionedgrouping scheme, MUEs 1 and MUEs 2 will be consid-ered as cell-centre MUEs and cell-edge MUEs, respec-tively. Then, depending on the target scheme, either cell-edge or cell-centre MUEs are first scheduled with RNs inUn subframes in such a way that the resource fairness isassured. It is to be noted that co-scheduling cell-edge orcell-centre MUEs has different impacts on the interferencecharacteristics during Uu and Un subframes. This is as welldemonstrated in Figure 13 for 10-RN urban deploymentutilising FCPC. In this figure, the changes in the mean IoTlevels due to the MUEs selection scheme are depicted bymean IoT indicators where a pointer direction towards zeroimplies no change in the mean IoT level with respect to therandom MUEs selection scheme. Furthermore, a pointerdirection towards plus or minus indicate a relative increaseor a decrease in the mean IoT level, respectively.As also depicted by the mean IoT indicators, for the cell-

centre MUEs selection scheme, the total mean IoT atDeNB increases during Uu subframes as cell-edge MUEsare dominating the inter-cell interference, whereas it


45

Ö. Bulakci et al.

Table VI. Performance of flexible resource split at donor evolved node B in suburban scenarios.

MUE Scenario TP gain (%) P0 (dBm)selection over eNB-only

scheme 90th 50th 5th MUE MUE RNpercentile percentile percentile (Uu) (Un)

Random 4 RNs 19 180 449 �63 �59 �6310 RNs 19 242 1881 �63 �55 �55

Cell 4 RNs 35 215 371 �59 �59 �63edge 10 RNs 28 253 1529 �59 �55 �55TP, throughput; Mn, macrocell-served user equipment; RN, relay node; eNB, evolved node B;(Uu), during Uu subframes; (Un), during Un subframes.

Mean IoT

Uu

+_ 0

RUE MUE

+_ 0

Total

RN MUE

TotalTooTT ttal_ 0 +_ 0 +_ 0+_ 0

Mean IoT

Uu

+_ 0

RUE MUE

+_ 0

Total

RN MUE

Total

+_ 0 +_ 0 +TooTT tal

_ 0+_ 0TooTT ttal

TooTT ttall

@

@@Un

@Un

+

Figure 13. Impact of macrocell-served user equipment (MUEs)selection schemes on user equipment (UEs) throughput (TP)cumulative distribution function (CDF) in flexible resource splitshown for 10-relay-node (RN) deployment utilising full compen-sation power control (FCPC); urban scenario. Res., resource;Sel., selection; RUEs, RN-served UEs; IoT, interference-over-thermal; eNB, evolved node B; @Uu, during Uu subframes;

@Un, during Un subframes.

decreases during Un subframes as cell-centre MUEs causeless interference. Accordingly, this scheme enhances theperformance of co-scheduled cell-centre MUEs and RNsyielding a gain at high TP regime; however, it decreasesthe performance of cell-edge MUEs yielding a loss at lowTP regime (see Cell centre MUEs Selection Scheme inTable V).In case of the cell-edge MUEs selection scheme, by con-

trast, the total mean IoT at DeNB increases during Unsubframes due to co-scheduled cell-edge MUEs, and itdecreases during Uu subframes as the cell-centre MUEsare dominating. Note that cell-centre MUEs cause lessinter-cell interference. Crucially, PC parameters of theMUE are updated to adapt to the changed interferencecharacteristics as shown in Table V (Cell-Edge MUEsSelection Scheme). Then, a gain in mid TP regime isobserved at the cost of a loss at the low UEs TP CDF per-centiles. Similar results are as well obtained in other urbanscenarios (Cell-Edge MUEs Selection Scheme in Table V).In particular, it is seen that cell-centre MUEs selectionresults in high loss at the 5th percentile UEs TP CDF level,yet improves the high TP regime. Note that a PC update is

seen necessary only if it yields a good trade-off betweendifferent TP regimes.Next, the impact of the MUEs selection for co-

scheduling is analysed in suburban scenarios (Figure 12;Table VI). Unlike urban scenarios, the suburban scenariosare not interference limited because of large ISD, and theperformance is mainly defined by the received signal-to-noise ratio. Consequently, power limited cell-edge MUEsshape the low TP regime. Therefore, the path loss of aMUEs is taken as the criterion to classify it as a cell-edgeor cell-centre MUEs. That is, the MUEs experiencing thelargest path losses, which are mainly the power-limitedUEs, are selected first for co-scheduling in the cell-edgeMUEs selection scheme. As the cell-centre MUEs selec-tion scheme does not provide a good trade-off betweenhigh and low TP regimes, we focus on the cell-edge MUEsselection scheme in the following. When the cell-edgeMUEs selection scheme is employed, the performance canbe further improved above the 25th percentile and 7th per-centile UEs TP CDF levels for 4-RN and 10-RN deploy-ments, respectively, at the cost of reduced performance atlower TP percentiles. Crucially, the cell-edge MUEs selec-tion scheme enables an increased PMUE

0 value during Uusubframes, which is the main reason for the performanceenhancement (see Cell-Edge MUEs Selection Scheme inTable VI). We note that such an increase in P0 value isotherwise not desirable because of power limitation of cell-edge MUEs. It is also worth noting that the PMUE

0 settingof the co-scheduled MUEs should be set differently thanthat of the other MUEs in most of the considered cases insuburban scenarios.

6. CONCLUSION

In this work, we have investigated different resource shar-ing techniques taking into account UL PC optimisation. Wehave proposed a combination of resource allocation on therelay link based on the number of attached RUEs and aTP throttling scheme achieving MMF in the end-to-endtwo-hop communication. Furthermore, we have demon-strated a flexible resource split at DeNB based on over-provisioning of Un subframes and co-scheduling of MUEs,


Ö. Bulakci et al.

which enables a fast adaptation to dynamic system con-ditions and enhances the resource utilisation efficiency atDeNB. Thorough system-level simulations within the LTE-Advanced framework have been carried out in 3GPP urban(Case 1—ISD 500 m) and suburban (Case 3—ISD 1732 m)scenarios considering 4 and 10 RNs per cell. Results showsignificant TP enhancements and high fairness gains inboth RN cells and in the system as a whole at the cost of anegligible decrease in cell capacity, when the hard resourcesplit is used at DeNB. It is shown that the system perfor-mance can be further enhanced, when the flexible resourcesplit is utilised at DeNB. In particular, results reveal highTP gains in a wide range of user TP CDF levels. In thelatter scheme, two MUEs selection schemes are provided,where various trade-offs between low and high TP regimescan be attained.

REFERENCES

1. Bou Saleh A, Redana S, Hämäläinen J, Raaf B. On thecoverage extension and capacity enhancement of inbandrelay deployments in LTE-Advanced networks. Journalof Electrical and Computer Engineering 2010; 2010:1–12. Article ID 894846.

2. Schoenen R, Zirwas W, Walke BH. Capacity and cover-age analysis of a 3GPP-LTE multihop deployment sce-nario. In ICC Workshops 2008, IEEE: Beijing, China,19–23 May 2008; 31–36.

3. Venkatkumar V, Haustein T, Faulkner M. Relayingresults for indoor coverage in long-term evolution andbeyond. European Transactions on Telecommunications2010; 21: 770–779.

4. Lang E, Redana S, Raaf B. Business impact of relaydeployment for coverage extension in 3GPP LTE-Advanced. In ICC Workshops 2009, IEEE: Dresden,Germany, 14–18 June 2009; 1–5.

5. Priyanto BE, Sorensen TB, Jensen OK. In-band inter-ference effects on UTRA LTE uplink resource blockallocation. In VTC 2008—Spring, IEEE, Marina Bay,Singapore, 11–14 May 2008; 1846–1850.

6. Bulakci Ö, Redana S, Hämäläinen J, Raaf B. Systemoptimization in relay enhanced LTE-Advanced networksvia uplink power control. In VTC 2010—Spring, IEEE,Taipei, Taiwan, 16–19 May 2010; 1–5.

7. Bulakci Ö, Redana S, Raaf B, Hämäläinen J. Impact ofpower control optimization on the system performanceof relay based LTE-Advanced heterogeneous networks.Journal of Communications and Networks, Special Issueon Heterogeneous Networks 2011; 13(4): 345–359.

8. Bulakci Ö, Bou Saleh A, Redana S, Raaf B,Hämäläinen, J. Flexible backhaul resource sharing anduplink power control optimization in LTE-Advancedrelay networks. In VTC 2011—Fall, IEEE, SanFrancisco, USA, 5–8 September 2011; 1–6.

9. Roth S, Gan J, Danev D. Subframe allocation for relaynetworks in the LTE-Advanced standard. In PIMRC2010, IEEE, Istanbul, Turkey, 26–29 September 2010;1758–1763.

10. Jing H, Haiming W. Uplink performance evaluation ofwireless self-backhauling relay in LTE-Advanced. InWiCOM 2010, IEEE, Chengdu, China, 23–25 September2010; 1–4.

11. Wei H, Jing H, Haiming W. Full uplink performanceevaluation of FDD/TDD LTE-Advanced networks withtype-1 relays. In VTC 2011—Fall, IEEE, San Francisco,USA, 5–8 September 2011; 1–5.

12. TS 36.300. Evolved Universal Terrestrial Radio Access(E-UTRA) and Evolved Universal Terrestrial RadioAccess Network (E-UTRAN); Overall description;Stage 2 (Release 10). 3GPP, v.10.8.0, June 2012.

13. Ciochina C, Mottier D, Sari H. An analysis of three mul-tiple access techniques for the uplink of future cellularmobile systems. European Transactions on Telecommu-nications 2008; 19: 581–588.

14. TR 36.814. Evolved Universal Terrestrial Radio Access(E-UTRA); Further advancements for E-UTRA phys-ical layer aspects (Release 9). 3GPP, v. 9.0.0, March2010.

15. TS 36.213. Evolved Universal Terrestrial Radio Access(E-UTRA); Physical layer procedures (Release 8).3GPP, v.8.8.0, September 2009.

16. TS 36.101. Evolved Universal Terrestrial Radio Access(E-UTRA); User Equipment (UE) radio transmissionand reception (Release 8). 3GPP, v.8.18.0, June 2012.

17. Castellanos CU, Villa DL, Rosa C, Pedersen KI,Calabrese FD, Michaelsen P-H, Michel J. Performanceof uplink fractional power control in UTRAN LTE. InVTC 2008-Spring, IEEE, Marina Bay, Singapore, 11–14May 2008; 2517–2521.

18. Simonsson A, Furuskar A. Uplink power control inLTE—overview and performance. In VTC 2008-Fall,IEEE, Calgary, Canada, 21–24 September 2008; 1–5.

19. Müllner R, Ball CF, Boussif M, Lienhart J, Hric P,Winkler H, Kremnitzer K, Kronlachner R. Enhancinguplink performance in UTRAN LTE networks by loadadaptive power control. European Transactions onTelecommunications 2010; 21: 458–468.

20. Holma H, Toskala A. LTE for UMTS—Evolution toLTE-Advanced (2nd edn). Wiley, 2011.

21. Laselva D, Capozzi F, Frederiksen F, Pedersen KI,Wigard J, Kovacs IZ. On the impact of realistic con-trol channel constraints on QoS provisioning in UTRANLTE. In VTC 2009—Fall, IEEE, Alaska, USA, Septem-ber 2009; 1–5.

22. Viering I, Lobinger A, Stefanski S. Efficient uplink mod-eling for dynamic system-level simulations of cellular


47

Ö. Bulakci et al.

and mobile networks. EURASIP Journal on WirelessCommunications and Networking 2010; 2010. ArticleID 282465.

23. Calabrese FD, Rosa C, Anas M, Michaelsen PH,Pedersen KI, Mogensen PE. Adaptive transmissionbandwidth based packet scheduling for LTE uplink. InVTC 2008—Fall, IEEE, Calgary, Canada, September2008; 1–5.

24. R1-091294. On the design of relay node for LTE-Advanced. 3GPP, Texas Instruments, March 2009.

25. Mogensen P, Wei N, Kovacs IZ. LTE capacity comparedto the Shannon bound. In VTC 2007—Spring, IEEE,Dublin, Ireland, 22–25 April 2007; 1234–1238.

26. Bou Saleh A, Bulakci Ö, Redana S, Raaf B,Hämäläinen J. A divide-and-conquer approach tomitigate relay-to-relay interference. In PIMRC 2011,

IEEE, Toronto, Canada, 11–14 September 2011;1889–1893.

27. Bulakci Ö, Nedelcu AS, Bou Saleh A, RedanaS, Hämäläinen J. Impact of backhaul subframemisalignment on uplink system performance ofLTE-Advanced relay networks. In VTC 2012-Fall, IEEE, Québec City, Canada, 3–6 September2012; 6.

28. Jain RK, Chiu DMW, Hawe WR. A quantitative mea-sure of fairness and discrimination for resource allo-cation in shared computer systems. Public, TR-301,Digital Equipment Corp., 26 September 1984.

29. Himayat N, Talwar S, Rao A, Soni R. Interferencemanagement for 4G cellular standards [WiMAX/LTEUPDATE]. Communications Magazine, IEEE 2010;48(8): 86–92.


Publication VII - Aalto

Documents

Transcript of Publication VII - Aalto